| Prev | Chapter�6.�Production Environments | Next |
|---|
Did performance problems happen overnight, or did they sneak up on +
Did performance problems happen overnight, or did they sneak up on you? Any clue what caused the performance problems (e.g. loading 20K users into .LRN)
Is the file system out of space? Is the machine swapping to disk constantly?
Isolating and solving database problems.
Without daily internal maintenance, most databases slowly degrade in performance. For PostGreSQL, see the section called “Vacuum Postgres nightly”. For Oracle, use exec dbms_stats.gather_schema_stats('SCHEMA_NAME') (Andrew Piskorski's Oracle notes).
You can track the exact amount of time each database query on a page takes:
Go to Main Site : Site-Wide Administration : Install Software
Click on "Install New Application" in "Install from OpenACS Repository"
Choose "ACS Developer Support">
After install is complete, restart the server.
Browse to Developer Support, which is automatically mounted at /ds. -
Turn on Database statistics
Browse directly to a slow page and click "Request Information" at the bottom of the page.
This should return a list of database queries on the page, including the exact query (so it can be cut-paste into psql or oracle) and the time each query took.
Identify a runaway Oracle query: first, use ps aux or top to get the UNIX process ID of a runaway Oracle process.
Log in to SQL*Plus as the admin:
[$OPENACS_SERVICE_NAME ~]$ svrmgrl
+ Turn on Database statistics
Browse directly to a slow page and click "Request Information" at the bottom of the page.
This should return a list of database queries on the page, including the exact query (so it can be cut-paste into psql or oracle) and the time each query took.
Identify a runaway Oracle query: first, use ps aux or top to get the UNIX process ID of a runaway Oracle process.
Log in to SQL*Plus as the admin:
[$OPENACS_SERVICE_NAME ~]$ svrmgrl
Oracle Server Manager Release 3.1.7.0.0 - Production
@@ -26,7 +26,22 @@
where sql.address = s.sql_address
and sql.hash_value = s.sql_hash_value
--and upper(s.username) like 'USERNAME%'
- order by s.username ,s.sid ,s.serial# ,sql.piece ;To kill a troubled process:
alter system kill session 'SID,SERIAL#'; --substitute values for SID and SERIAL#
+ order by s.username ,s.sid ,s.serial# ,sql.piece ;
To kill a troubled process:
alter system kill session 'SID,SERIAL#'; --substitute values for SID and SERIAL#
(See Andrew Piskorski's Oracle notes)
Identify a runaway Postgres query. First, logging must be enabled in the database. This imposes a performance penalty and should not be done in normal operation.
Edit the file postgresql.conf - its location depends on the PostGreSQL installation - and change
#stats_command_string = false
to
stats_command_string = true
Next, connect to postgres (psql service0) and select * from pg_stat_activity;. Typical output should look like:
+ datid | datname | procpid | usesysid | usename | current_query +----------+-------------+---------+----------+---------+----------------- + 64344418 | openacs.org | 14122 | 101 | nsadmin | <IDLE> + 64344418 | openacs.org | 14123 | 101 | nsadmin | + delete + from acs_mail_lite_queue + where message_id = '2478608'; + 64344418 | openacs.org | 14124 | 101 | nsadmin | <IDLE> + 64344418 | openacs.org | 14137 | 101 | nsadmin | <IDLE> + 64344418 | openacs.org | 14139 | 101 | nsadmin | <IDLE> + 64344418 | openacs.org | 14309 | 101 | nsadmin | <IDLE> + 64344418 | openacs.org | 14311 | 101 | nsadmin | <IDLE> + 64344418 | openacs.org | 14549 | 101 | nsadmin | <IDLE> +(8 rows) +openacs.org=>
The first task is to create an appropriate environment for finding out what is going on inside Oracle. Oracle provides Statspack, a package to monitor and save the state of the v$ performance views. These reports @@ -43,7 +58,7 @@
To be able to get a overview of how Oracle executes a particular query, install "autotrace". I usually follow the instructions here http://asktom.oracle.com/~tkyte/article1/autotrace.html. -
+